WIP: GRT: Add detour metric tracking for routing stages (#8941) by Divinesoumyadip · Pull Request #9641 · The-OpenROAD-Project/OpenROAD

Divinesoumyadip · 2026-03-05T07:56:01Z

Hey @eder-matheus , opening this as a draft PR to get the initial architecture up. I've wired up the full Tcl -> C++ pipeline based on your feedback.

What's working so far:

Added the report_net_global_routing_detour -nets <net_names> Tcl command (wildcard support is built-in via get_nets).

Wired the SWIG bindings and C++ headers directly to GlobalRouter.

 Implemented the core metric functions in FastRouteCore (computeTotalWirelength, computeNetSttWirelength, computeNetFinalWirelength).

 Added the stt_wirelengths_ map to track STT wirelength per net.

Still left to do:

Hook up the internal FastRoute netID mapping to the new GlobalRouter::reportNetGlobalRoutingDetour function.

Clean up and format the final logger output.

Let me know if this structural direction looks good to you before I finalize the internal variable mapping!

gemini-code-assist

Code Review

This pull request introduces functionality for tracking detour metrics in the global router. The changes include a new Tcl command, C++ bindings, and core metric computation functions. The overall structure is sound and the new functionality is well-integrated. I have one suggestion to refactor a newly added function to improve code reuse and maintainability, aligning with our guidelines on avoiding code duplication.

_{Note: Security Review did not run due to the size of the PR.}

gemini-code-assist · 2026-03-05T07:58:24Z

src/grt/src/fastroute/src/maze.cpp

+  int total = 0;
+  for (const int& netID : net_ids_) {
+    const auto& treeedges = sttrees_[netID].edges;
+    const int num_edges = sttrees_[netID].num_edges();
+    for (int edgeID = 0; edgeID < num_edges; edgeID++) {
+      const auto& edge = treeedges[edgeID];
+      if (edge.len > 0) {
+        total += edge.route.routelen;
+      }
+    }
+  }
+  return total;


The logic within computeTotalWirelength duplicates the functionality of computeNetFinalWirelength. To enhance maintainability and reduce code duplication, computeTotalWirelength can be refactored to iterate through the nets and call computeNetFinalWirelength for each one.

int total = 0; for (const int netID : net_ids_) { total += computeNetFinalWirelength(netID); } return total;

References

Encapsulate the logic for calculating master symmetry in a helper function to avoid code duplication.

eder-matheus · 2026-03-05T16:00:23Z

src/grt/src/GlobalRouter.tcl

+
+  set block [ord::get_db_block]
+  if { $block == "NULL" } {
+    utl::error GRT 240 "Missing dbBlock."


Notice that the message ID 240 and 241 are already used in the code. You can check the file OpenROAD/src/grt/messages.txt to see all the IDs already used.

eder-matheus · 2026-03-05T16:05:22Z

src/grt/src/GlobalRouter.cpp

  if (!nets.empty()) {
    routes = fastroute_->run();
+    int wl_after_routing = fastroute_->computeTotalWirelength();
+    logger_->info(GRT, 300, "Total wirelength after routing: {} tiles", wl_after_routing);


Why this message is necessary? We already report the wire length at the end of the routing flow (check the computeWirelength function).

I see that your new computeTotalWirelength uses the internal FastRouteCore unit. This is good for debug purposes (which is the goal of the detour metric feature), but not really useful for users to have it as a default message.

eder-matheus · 2026-03-05T16:05:54Z

src/grt/src/fastroute/src/maze.cpp

+  return total;
+}
 }  // namespace grt
+


Remove the extra lines (here and in other files where it might happens).

eder-matheus · 2026-03-05T16:15:57Z

src/grt/src/fastroute/src/maze.cpp

+int FastRouteCore::computeNetSttWirelength(int netID)
+{
+  int total = 0;
+  for (const auto& edge : sttrees_[netID].edges) {
+    total += edge.len;
+  }
+  return total;
+}


In this function, the sttrees_ name is misleading. What you really want is to track the wirelength computed inside RSMT.cpp file, at the gen_brk_RSMT function. There we'll get the Steiner tree directly from Flute/PD, which is the starting point of the routing topology.

eder-matheus · 2026-03-05T16:22:26Z

src/grt/src/fastroute/src/maze.cpp

+int FastRouteCore::computeNetFinalWirelength(int netID)
+{
+  int total = 0;
+  for (const auto& edge : sttrees_[netID].edges) {
+    if (edge.len > 0) {
+      total += edge.route.routelen;
+    }
+  }
+  return total;
+}


You can use this function for general purposes. We want to track the wire length at 4 main stages:

After Steiner tree construction from Flute/PD

After building the Rectilinear Steiner Tree, before congestion iterations

Rectilinear Steiner Tree, after congestion iterations

Rectilinear Steiner Tree, after layer assignment and 3D maze route

You can check where in the code each of these steps happen by searching for the steinerTreeVisualization and StTreeVisualization functions. They are used for debug on the GUI, and are called at the locations where we want to track WL.

I think this function can be used to track the WL on stages 2, 3 and 4, and the computeNetSttWirelength can be used for 1.

eder-matheus · 2026-03-05T16:24:53Z

@Divinesoumyadip DCO is failing in the PR. You need to sign-off your commits by adding the -s flag to the git commit call: git commit -s -m "<your commit message>".

Check the DCO link in the checks for advise on how to fix DCO.

Divinesoumyadip · 2026-03-05T17:31:03Z

@Divinesoumyadip DCO is failing in the PR. You need to sign-off your commits by adding the -s flag to the git commit call: git commit -s -m "<your commit message>".

Check the DCO link in the checks for advise on how to fix DCO.

Thanks for the incredibly detailed review, Eder! This clarifies the internal topology flow perfectly.

I will:

Update the message IDs in GlobalRouter.tcl to unused ones from messages.txt.
Strip out the redundant total wirelength logger in GlobalRouter.cpp.
Clean up the trailing whitespaces.
Remap the tracking stages exactly as you described: I'll hook computeNetSttWirelength into gen_brk_RSMT in RSMT.cpp for the initial Flute/PD tree, and then inject computeNetFinalWirelength at the steinerTreeVisualization call sites to capture stages 2, 3, and 4.

I will get these changes implemented, sign-off the commits to clear the DCO, and push the update shortly

Divinesoumyadip · 2026-03-05T18:13:54Z

@Divinesoumyadip DCO is failing in the PR. You need to sign-off your commits by adding the -s flag to the git commit call: git commit -s -m "<your commit message>".

Check the DCO link in the checks for advise on how to fix DCO.

Hi @eder-matheus , I have now amended the commits with the --signoff flag to satisfy the DCO requirement. I also included the GPoint3D forward declaration in GridGraph.h to fix the Jenkins build error. Everything should be green once the current build finishes.

- Add computeTotalWirelength() to FastRouteCore in maze.cpp - Add public declaration in FastRoute.h - Call after fastroute_->run() in GlobalRouter::findRouting() Part of issue The-OpenROAD-Project#8941: GRT detour metric implementation Signed-off-by: Divinesoumyadip <soumyadipdasmahapatra343@gmail.com>

- Add computeNetSttWirelength() to get STT wirelength per net - Add computeNetFinalWirelength() to get final wirelength per net - Add stt_wirelengths_ map to store per-net STT wirelength - Add getSttWirelengths() accessor Part of issue The-OpenROAD-Project#8941 Signed-off-by: Divinesoumyadip <soumyadipdasmahapatra343@gmail.com>

Signed-off-by: Divinesoumyadip <soumyadipdasmahapatra343@gmail.com>

eder-matheus

Most of the changes in the code are related to comments being removed. I can't see a reason for it, so please undo these changes and make sure new updates will stay on the scope of the PR.

eder-matheus · 2026-03-06T17:48:32Z

src/grt/src/fastroute/src/maze.cpp

  const auto& treeedges = sttrees_[net_id].edges;
  const int num_edges = sttrees_[net_id].num_edges();

-  // group all edges that crosses the same position


Why removing this and the other comments? They don't relate to the goal of this PR

eder-matheus · 2026-03-06T17:50:04Z

src/grt/src/fastroute/src/maze.cpp

+  }
+
+  for (const int netID : net_ids_) {
+    final_wirelengths_[netID] = computeNetFinalWirelength(netID);


final_wirelengths_ does not exists in the main class, causing a build error.

gemini-code-assist bot reviewed Mar 5, 2026

View reviewed changes

Divinesoumyadip mentioned this pull request Mar 5, 2026

grt: add incremental routing support to CUGR #9645

Closed

maliberty requested a review from eder-matheus March 5, 2026 15:56

eder-matheus reviewed Mar 5, 2026

View reviewed changes

Divinesoumyadip force-pushed the detour-metric-issue-8941 branch from 29d6a86 to 95f904b Compare March 5, 2026 18:00

Divinesoumyadip added 4 commits March 6, 2026 12:49

GRT: add Tcl command report_net_global_routing_detour with C++ bindings

1179cce

Signed-off-by: Divinesoumyadip <soumyadipdasmahapatra343@gmail.com>

GRT: refactor computeTotalWirelength to reuse computeNetFinalWirelength

6203c83

Signed-off-by: Divinesoumyadip <soumyadipdasmahapatra343@gmail.com>

Divinesoumyadip force-pushed the detour-metric-issue-8941 branch from 95f904b to 6203c83 Compare March 6, 2026 07:19

eder-matheus requested changes Mar 6, 2026

View reviewed changes

Conversation

Divinesoumyadip commented Mar 5, 2026

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus commented Mar 5, 2026

Uh oh!

Divinesoumyadip commented Mar 5, 2026

Uh oh!

Divinesoumyadip commented Mar 5, 2026

Uh oh!

eder-matheus left a comment

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

eder-matheus Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants